Translation Issues 4-97 (December 1997) by Daniel Brockmann

a Perfect Match to Save Translation Cost

TC Forum 2/97 featured a very interesting article by Lehrndorfer/Mangold. In their article they focused on saving translation cost by using controlled language in combination with machine translation systems. In the present article, I would like to extend their approach to a recent phenomenon in the area of computer-assisted translation: the increasingly popular translation memory systems.

1. I will briefly sketch the architecture of such systems, using MultiTerm and Translator's Workbench by TRADOS as an example. I will give examples showing the positive impact of controlled language on the use of such systems.
2. I would like toill give a brief perspective on future developments.

It goes without saying that controlled language makes it easier not only to understand a text, but also to translate it into another language, thereby reducing translation cost. This positive effect can be even more increased by the use of professional translation tools. By "translation tools", I do not mean machine translation systems such as Logos or Systran, but rather terminology database and translation memory applications. Typical examples of such tools are MultiTerm '95 Plus and Translator's Workbench.

These systems assist the professional translator on three levels:

• terminology (active terminology recognition),
• sentence part (text search in translation memory), and
• sentence level (finding identical or similar sentences in translation memory, using fuzzy-matching technology).

To illustrate this, let's use some of the examples cited by Lehrndorfer/Mangold in their TCF 2/97 article. Let's assume a translator has to translate the sentence "Nie Magnesium-Teile mit cyanidisch verzinkten Teilen kombinieren". Let's furthermore suppose that the terminology database contains entries for the terms "Magnesium-Teil" (magnesium part), "cyanidisch verzinkt" (cyanide plated), "verzinkt" (plated), and "rostbeständiger Stahl" (rustproof steel). For the sentence above, the active terminology recognition will find the German terms along with their English equivalents. The translator can paste them into the target text with a few mouse clicks:

As a consequence, the effort going into the translation of this sentence is reduced considerably, thanks to the known terminology that the translator can use immediately for his or her translation.

Let's now assume that the translator comes across the sentence "Nie verzinkte Bauteile mit rostbeständigen Stählen kombinieren" at a later stage. Thanks to controlled language, this sentence is similar to the one he or she has already translated above. As a result, the translation memory will find a fuzzy match and generate a proposal based on the previous translation. The translator needs to adapt some terms in the new translation. However, since these terms are all in the terminology database, it only takes two mouse clicks to complete the new translation:

In conclusion, controlled language not only makes life easier with machine translation systems whose usefulness is still controversial, given the more or less tedious pre- and post-editing tasks. It also greatly enhances the use of translation memory and terminology tools. The more controlled a source text, the more efficient these tools will be in the translation process. In the medium term, they will also be adapted for source-text authoring. This means that the writer will be able to re-use his or her own material using an "authoring memory", thus increasing consistency even more in the source language.  

"transline Deutschland - Übersetzungsdienst für technische Übersetzung"